What if your insurance company could decide your claim without ever seeing it?

Not by policy. Not by permission. By math.

You had a car accident. You file a claim. Today, every adjuster, every contractor, every fraud analyst reads everything — your damage estimate, your fault assessment, your medical details, your entire claims history. It gets forwarded, stored, sometimes breached.

In this demo, you'll file a claim. Your data gets encrypted on your device. The server that processes it has no secret key and is mathematically incapable of reading your data. It runs six policy gates on encrypted values — coverage, deductible, fault, fraud, injury, jurisdiction — and produces a binding decision. Approve, deny, or flag. Route to fast-track, standard review, or special investigation. Set a payout band. All without seeing a single number.

ⓘInteractive demo — runs entirely in your browser. The encryption and hashes are computed live with real WebCrypto; the policy gates, server boundary, and post-quantum receipt shown here are an illustrative walkthrough, not a live production server call. In production, these AI decisions are governed by Agent-008, and the 74-byte evidence is produced by H33-74.

1

Your claim

Pick any numbers. This prevents the demo from being pre-generated.

2

Only ciphertext leaves your device

Your claim was encrypted here. The server received scrambled bytes. It cannot read them — even if it wanted to.

🔐The server is mathematically incapable of decryption. It has no secret key and no decrypt method. This is structural — built into the code, not a setting.

server boundary

3

The server knows outcomes. It does not know inputs.

Six policy gates ran on your encrypted claim. Each produced a single bit: pass or fail. The server never saw the values behind them.

Policy Gate Results

Routing Decision

This decision was made on data the system cannot see.

🛠This is not a recommendation. This is an enforcement action. The routing is binding. The payout band is set. No adjuster read your claim to get here.

confidence boundary

CONFIDENCE BOUNDARY Mode: Threshold Proof Score Disclosure: None (pass/fail per gate only) Leakage: 1 bit per gate × 6 gates = 6 bits total The server knows which gates passed. It does not know: - the damage amount (only: above or below deductible) - the fault percentage (only: above or below 80%) - the claims count (only: above or below 3) - the injury details (only: injury present or not)

4

This proves it.

74 bytes. Binds the claim, the ciphertext, all six gate results, the routing, and the timestamp. Signed under three independent hardness assumptions. The insurer proves the decision was correct. You prove your data was never exposed.

attestation receipt

5

Independent verification

A separate verifier checks every binding. If anything was tampered with — your claim, the decision, the server binary — the proof fails. This is what a regulator, auditor, or reinsurer would see.

Try to break it

🎯In production, decisions like this are governed by Agent-008 — the authority under which AI decisions are made and bound to evidence. Related demos: Agent-008 governance · Agent-008 preflight · Root ceremony.

Learn More About Agent-Zero Schedule a Demo

H33-Agent-Zero Demo: Provable Encrypted Decisions